Brief
Ovenglint Home Appliances is an invented retailer of kitchen and laundry appliances selling online and through showrooms. Many customers research online, compare models, then buy in a showroom, so a large share of the sales that marketing influences never touches the website's checkout. Its ad platforms, its analytics property (GA4Google's event-based analytics platform.) and finance all report different revenue, consent rates differ sharply by device, and the board wants to know which channels actually drive sales.
The marketing director needs a strategy that answers the board's question honestly, stops the monthly argument about revenue and fits a modest budget and a small team. She does not want a tool list; she wants to know which decisions each piece of work will improve.
Inputs and constraints
Ovenglint Home Appliances: tracking audit and consent
Scored with the analytics implementation scorecard:
| Criterion | Rating | Evidence |
|---|---|---|
| Measurement plan (gate) | Partial | Written two years ago; showroom sales absent |
| Consent behaviour (gate) | Partial | One advertising tag fires before the visitor chooses |
| Key events (gate) | Partial | Purchase fires again when the confirmation page is refreshed |
| Campaign tagging | Partial | Email and affiliate links tagged inconsistently |
| Platform integrations | Partial | Ad accounts linked; no warehouse export |
| Reconciliation | Missing | Online purchases never compared with the order system |
Consent to analytics: about 70% of desktop visitors and 45% of mobile visitors.
Ovenglint Home Appliances: spend, revenue and team
| Channel | Annual spend |
|---|---|
| Paid search and Shopping | £720,000 |
| Paid social | £260,000 |
| Regional radio | £150,000 |
| Affiliates | £60,000 |
| Email (platform cost) | £40,000 |
Revenue for the same year: ad platforms' reported revenue added together £9.8m; analytics purchase revenue £5.1m; finance £14.2m, of which £6.0m online and £8.2m in showrooms.
- Weekly showroom sales by region from finance for three years; showrooms in nine regions.
- Team: one analyst who writes SQL, a marketing operations manager who owns Tag Manager, an agency running paid media; no data engineer.
- Data budget for tools and hosting: about £30,000 a year.
Constraints: no measurement change may break checkout; the agency must keep optimising during the work; any test must avoid the peak trading weeks around seasonal sales events.
Task
Produce one strategy document with these sections, in order. Give a reason for every tool and every method, tied to a decision.
- Baseline. Confirm the scoring with the analytics implementation scorecard, name the failed gates, and list the fixes that must happen before any new layer is built.
- Architecture. Trace data from collection to decision: consent and tags, the BigQuery export and what it is for, how online orders, showroom sales, spend and CRM records reach the warehouse, and whether server-side tagging earns its cost at this team size. Draw it as a table of stage, component, owner and the decision it serves.
- Method per decision. List the board's and the team's real decisions, from the annual channel split to weekly bid changes, and assign attribution, incrementality tests or mix modelling to each, using the guide to choosing an attribution approach and the comparison of mix modelling and multi-touch attribution. Say what each method cannot see.
- First test. Design one test following the incrementality testing guide: the channel and why it goes first, units, metric including showroom sales, minimum detectable effect, guardrails, duration and the decision rule. Record it in the experiment hypothesis template.
- Governed metrics. Write definitions for online revenue, showroom revenue, total revenue, orders, new customers and marketing cost, with formula, grain, filters, time rule, owner and source of truth, following the metrics layer guide. Name every KPIA metric tied directly to an objective. the board pack will carry and which definition it reads.
- Reporting and governance. Say which dashboards are certified, who owns them and how they are reviewed and retired, using the dashboard governance guide.
- Roadmap. Stage the work over the next year or so by value and capability, with an owner and a review point for each stage, and say when a mix model (MMMStatistical model of channel impact.) becomes worth building and what calibrates it.
- Board summary. One page: what the board will be able to answer, when, and with what confidence.
Evaluation criteria
| Criterion | What a strong answer shows | Common gap |
|---|---|---|
| Architecture | Traces collection, consent, server-side options and the warehouse with reasons for each choice | Adds tools without saying what decision they serve |
| Method per decision | Assigns attribution, incrementality tests and mix modelling to named decisions within the data available | Picks one method for everything |
| First test | Designs a geo or holdout test with metric, guardrails, duration and decision rule | A test with no decision rule |
| Governed metrics | Writes definitions with formula, grain, filters, owner and source of truth | Leaves revenue undefined across teams |
| Roadmap | Stages the work by value and capability with owners and review points | A wish list with no sequence |
A strong strategy fixes the gates before building anything on them, because a warehouse full of duplicated purchases only makes the wrong number faster. It treats the revenue gap as three different questions rather than one error: platforms credit what they touched, analytics sees consented online sessions, and finance sees everything including showrooms. It chooses a geo design for the first test because only regional showroom sales can show what media does to sales that never reach the checkout, and it reads that test as an IncrementalityConversions caused by marketing. factor for routine reporting and a calibration point for the later model. It recognises that the existing weekly showroom history by region is the asset that makes both geo tests and a mix model possible.
Check your reading of the inputs against the figures below, after working them out yourself.
Reference figures for self-checking
- Ad platforms claim £9.8m against £6.0m of online revenue in finance: they over-claim online, and none of them can see the £8.2m in showrooms directly.
- Analytics sees £5.1m of £6.0m online, about 85%, consistent with consent gaps and the duplicated purchase in opposite directions; the reconciliation has to separate them.
- Showrooms carry about 58% of revenue (£8.2m of £14.2m), so any method limited to online paths misses most of what the board is asking about.
- Total spend in the table: £1.23m, about 9% of revenue.
Submission checklist
When you have finished, compare your reconciliation thinking with the BigQuery reconciliation exercise for a soap maker, your test with the geo holdout exercise for a juice delivery firm and your definitions with the metric definitions exercise for a training provider. For the measurement plan and dashboard foundations this strategy builds on, revisit the intermediate analytics level.
